Recommendation systems are the most successful and widespread applications of Machine Learning. These systems filter out products to the users based on the previous data involved with them. If a user doesn’t have any previous history then, in this case, the popular products are recommended. These systems are integrated with most of the tech giants we see today including Amazon, Netflix, LinkedIn to help discover users new products, movies and shows, jobs etc.
Recommendation systems are generally classified into Content Based Filtering and Collaborative based Filtering. Both these techniques are unique in themselves and are based on the user-product interactive behavior patterns. Further these two models combined together constitutes Hybrid Based Recommendations. Also, there exists different models under such domain following alternative strategies but for now in this blog we will study Collaborative and Content based Recommendation Systems.
1. Content Based Filtering
The content based filtering model captures the feedback and previous actions of a user and recommends new similar products to them. The parameters on which this model works, is hand featured by the user’s actions itself. The user’s previous historical data is fed to the model which further acts as a recommendation system to them.
For instance, let’s consider recommendation systems under the movies domain. We have movies of different genres, namely — Adventurous, Horror, Sci-fi, Romance, Thriller, etc.
Suppose a user U1 watches a movie M1 under the adventurous genre. He likes the movie and gives a rating of 5/5 for it. Then U1 watches another movie M2 of the same genre and gives a rating of 4/5. So we see that both these movies come under the same genre or under the same content i.e. Adventurous. Here the point to be considered is content.
Moving forward, if we consider these two movies, M1 and M2 which are under the same content, then a new movie or an existing movie suppose M3 of genre — adventurous whose parameters —Actors, Director, Production Team, etc. are common to that of movies M1 and M2, will be recommended to the user U1. This is how Content based filtering works in the recommendation system.
- Since the content based filtering model is specific to a single user as in recommendations, hence the model does not require data about any other users.
- Since the model is based on one single user, it recommends items specific to that user itself which other users may or may not be interested in.
- User satisfaction because recommendation is based on user interests.
- Since the model makes recommendations based on a single user, to other users this model won’t be of any use.
- Self customized user-product interactions present in this model, are limited only to that user.
- New products will not be recommended to the user as long as they don’t try something different.
2. Collaborative Based Filtering
Collaborative filtering is a technique used to filter out products and assign them to multiple users. This is an effective method to recommend new areas of interest to users which acts as a loophole in the content based filtering method. Generally this way of recommendation is helpful not only to a single user but to a set of users matching similar interest among the common features they possess.
Consider two users — U1 and U2 in the scenario. First user U1 watches movie M1 and gives a rating of 5/5 and user U2 also watches movie M1 and likes the movie and he too gives a rating of 5/5. Then U1 and U2 both again coincidentally watch another same movie M2 giving ratings 4/5 and 5/5 respectively. So both the users watch the same movies and give a high rating to both the movies.
Now, user U1 watches movie M3 and gives a rating of 4/5 which means he likes the movie. Note that here user U2 hasn’t watched movie M3. Now as per the user similarity and mentality towards the movies M1 and M2, M3 will be recommended to user U2. This takes place because of similar parameters (in this context — movie ratings) shown by both the users towards movies M1 and M2 which implies that there is a high possibility that user U2 likes the movie M3, hence movie M3 is recommended to user U2.
- Content based models expand their recommendation to new users.
- Models under this domain are trained on multiple user activity and recommendations of products are not limited to a single user.
- Models help users to discover new interests and areas outside their scope.
- There may exist a possibility of a user not liking a product recommendation.
- New items cannot be recommended if the user has not purchased or rated it.
- User might not be satisfied with all the features of the product recommended to them.
In collaborative based filtering, user similarity is taken into consideration whereas in Content based filtering, content similarity is the key.
Well I hope the blog was precise and clear with the concepts and features on the existing Recommendation System present in the industry.
Feel free to connect with me on these platforms —
- Mail : email@example.com
- LinkedIn : https://www.linkedin.com/in/aryan-kenchappagol/